Google DeepMind's AlphaGo platform recently won 60 online games of Go in a row, beating a slew of world champs. Here's what it means.
AlphaGo, Google DeepMind's machine learning platform, achieved a massive AI victory in March 2016 by defeating world champ Lee Sedol in Go, a game more complex than chess. But the computer brain isn't done: On Wednesday, a mysterious online Go player, dubbed "Master," revealed itself as the AI platform.
Playing the ancient Chinese game of Go on online gaming platforms Tygem and FoxGo, AlphaGo won 60 straight victories, many against top players from South Korea, China, and Japan.
The news hit a nerve for many, with one world champ, Gu Li, promising 100,000 yuan to anyone who could defeat AlphGo. As Fabio Cardenas, CEO of Sundown AI, put it, "AI's growing influence means that not even your hard-fought gaming status is safe."
AlphaGo won four out of five games in the live match against Sedol after mastering the game ten years earlier than many experts predicted. Several AI experts, including Marie desJardins, associate dean and professor of computer science at the University of Maryland, Baltimore County, Roman Yampolskiy, head of the Cybersecurity Lab at the University of Louisville, and Toby Walsh, AI professor at the University of New South Wales, put "money on the machine," especially since it had been trained specifically on Sedol's playing style. But it was nonetheless a major victory that an AI system could master such a complicated game so well. And with the news of the improved AlphaGo, "any doubts about AlphaGo being the best Go player in the world appear to have been put to rest," said Walsh.
AlphaGo's system, which uses convolutional neural networks, first worked on arcade games before moving on to Go. And while IBM's Deep Blue beat world chess master Gary Kasparov in 1997, that AI system was programmed specifically, rather than "teaching itself" through reinforcement learning as AlphaGo does. Go is also a highly complicated game, with potentially 200 options per move, compared to 20 on a chessboard—and experts say it relies heavily on intuition.
AI experts saw implications of the victory in other domains.
Given AlphaGo's previous success, Vincent Conitzer, computer science professor at Duke University, was not surprised by the news. But beyond AlphaGo's new achievement, Conitzer would be "more interested in progress in designing AI players for other complex games," and is "particularly interested in progress in games with imperfect information, where each player is only aware of part of the state of the game. This includes most card games and is also more representative of real-world strategic situations."
AlphaGo's improvement is "fascinating," said Manuela Veloso, head of machine learning at Carnegie Mellon University. Veloso also noted that she is curious "if and how AlphaGo's learning approach may apply to other different 'non-game' problems of other nature and relevance."
"It will be very interesting to see if human and computer Go players may now improve and learn by playing with AlphaGo," she said.
Walsh echoed this thought, noting that AlphGo's victories could even change the nature of the game, as it has "played moves that have surprised even Go masters."
"My expectation is that one of the biggest impacts may be that it opens up our understanding of the game, which is remarkable given that we've been playing Go for thousands of years," he said. "It will be interesting to see if, like computer chess, the best performance comes in the future from teams of humans and computers playing Go together."
"Will man and machine together be better than man or machine alone?" Walsh asked. "Each of us can bring unique strengths to the table."
- How Google's DeepMind beat the game of Go, which is even more complex than chess (TechRepublic)
- Can Google's DeepMind beat world Go champ? Watch live this week (TechRepublic)
- IBM Watson: The inside story of how the Jeopardy-winning supercomputer was born, and what it wants to do next (TechRepublic)
- Google AlphaGo AI clean sweeps European Go champion (ZDNet)
- Google AI gets better at 'seeing' the world by learning what to focus on (TechRepublic)
- How Google's AI breakthroughs are putting us on a path to narrow AI (TechRepublic)
- Smart machines are about to run the world: Here's how to prepare (TechRepublic)
- Google AI beats humans at more classic arcade games than ever before (TechRepublic)